The Future for Personalised Nutrition

When it comes to nutritional requirements, we have known for decades that one size does not fit all. Bigger and more physically active people need more energy than smaller or more sedentary people—an example of the physical law of conservation of energy. Bigger and more physically active people expend (lose) more energy so, to maintain equilibrium, they need to consume more dietary energy than do smaller or more sedentary people. The nutritional needs of pregnant or lactating women differ from those of non-pregnant and non-lactating women. This is well understood in terms of the nutrient needs for physiological processes such as the growth of the fetus, uterus and associated cells and tissues in pregnant women, the secretion of milk by breast-feeding women and the absence of iron loss through menstruation in pregnant women. Such well-established principles of nutritional physiology have underpinned the development of nutritional recommendations such as the Dietary Reference Values for Food Energy and Nutrients for the United Kingdom that provide recommendations stratified by age, sex and, in some cases, physiological state. The compilers of such dietary recommendations recognised that, even within a given population stratum, individuals differ in their nutritional needs. In most cases, such inter-individual variation follows a Gaussian distribution that has been operationalised, statistically, as three-point estimates of requirements. The Estimated Average Requirement (EAR) is an estimate of the mean requirement for the specific population. Two standard deviations above this is the Reference Nutrient Intake (RNI) that is estimated to cover the needs of almost all (97.5%) of the population group whereas two standard deviations below the EAR is the Lower Reference Nutrient Intake (LRNI)—an amount estimated to be sufficient for only a minority (2.5%) of the population. Whilst this strategy recognises important inter-individual differences, it offers no means of identifying who has higher or lower nutritional requirements and so the application of such recommendations is couched conservatively as being recommendations for groups of healthy people. The practice of dietetics starts with such population level recommendations and attempts to personalise them for specific individuals taking into account the individual's current dietary intake, health status, preferences and the desired nutritional trajectory needed to improve or maintain health. Inborn errors of metabolism (IEM) such as phenylketonuria (PKU), maple syrup urine disease, and galactosemia result from inherited or de novo mutations in genes encoding enzymes at key points in metabolic pathways that cause a buildup of toxic metabolites or failure to produce enough essential intermediates. Most individuals with IEM are diagnosed very soon after birth through routine screening involving biochemical and/or genetic testing. For example, individuals with PKU (who have limited ability to metabolise the amino acid phenylalanine because of mutations in the PAH gene that encodes phenylalanine hydroxylase) are prescribed a low protein diet with amino acid supplements designed to minimise exposure to phenylalanine. Whilst such relatively rare examples of personalised nutrition have been highly successful in practice, until about 25 years ago, there was no sustained attempt to understand the factors influencing individual nutritional requirements or inter-individual differences in response to diets, foods and nutrients in humans. It is a fundamental principle of genetics that genotype determines phenotype. Consequently, the success of the human genome mapping project and of downstream projects to map and characterise variants in genome sequence, principally single nucleotide polymorphisms (SNPs) and copy number variants (CNVs), offered a rational basis and experimental tools for investigating and understanding inter-individual differences in responses to diet and in nutritional needs. Of course, isolated examples of variants in nutrition-related genes had been known for much longer, and there was an emerging science of gene-nutrient interactions. For example, a common polymorphism in the MTHFR gene that encodes methylenetetrahydrofolate reductase (MTHFR) influences one-carbon metabolism. Individuals with a T instead of a C at position 677 in the MTHFR gene (rs1801133) have lower blood concentrations of folate and higher concentrations of homocysteine (Tsang et al. 2015) and require higher intakes of folate to normalise homocysteine concentrations. Post-genomic technological advances (initially using SNP-based microarrays but now more commonly sequencing techniques) made it possible to investigate huge numbers of variants across the genome of individuals simultaneously and to examine interactions between genotype and nutrition and their consequences for phenotype (so-called genome-wide association studies). Most of the studies in this area have been observational, which allow researchers to identify associations. To date, there have been rather fewer and usually much smaller, randomised controlled trials (RCTs) that provide evidence of causality (Caslake et al. 2008). For example, although there is good evidence that omega-3 fatty acids, for example, from fish oils, can have beneficial effects on blood lipids, a recent systematic review concluded that most of the evidence for gene-nutrient interactions in this area is weak (Keathley et al. 2022). However, the authors noted strong evidence for two specific gene-nutrient interactions (involving APOE-E4 carriers [rs429358, rs7412] and a 31-SNP nutrigenetic risk score, respectively) with omega-3 fatty acid intake, but generalisability was limited to specific populations (Keathley et al. 2022). Another recent systematic review reported considerable inconsistency in findings from studies of interactions between genetic variants and diet in relation to cardiovascular disease (CVD) risk (Roa-Díaz et al. 2022). Putative interaction effects were significant in some studies only and, even more worryingly, studies did not agree on the direction of effects (Roa-Díaz et al. 2022). Further, few reports of significant interactions were replicated, most studies lacked correction for multiple testing and often involved small sample sizes, which led the authors to conclude that the evidence for gene-diet interactions that influence CVD risk is limited (Roa-Díaz et al. 2022). In this Virtual Issue, Surendran and Vimaleswaran (2021) report findings from the GeNuIne (Gene-Nutrient Interactions) Collaboration on the effect of gene–nutrient interactions on vitamin B12 concentrations and cardio-metabolic disease risk factors in population-based studies from different ethnic groups. This is a welcome expansion of nutrigenetic research into more diverse communities but, as the authors note, further studies with larger sample sizes are needed to confirm or refute their findings. As part of the Food4Me Study (see below), Grimaldi et al. (2017) proposed a set of guidelines to evaluate scientific validity and evidence for genotype-based dietary advice. The intention was that this publication would stimulate debate on the utility of these guidelines leading to biennial revisions, as knowledge on the subject increased, but that intention has not been realised. Nearly 10 years later, there are still no agreed guidelines for the evaluation of personal genetic information to guide recommendations for dietary choices, an important element of strategies for personalised nutrition. In one of the earliest personalised nutrition RCTs involving healthy young Canadian adults, participants who were informed that they carried a risk variant of the ACE gene that increases sensitivity to the adverse effects of high salt intake and were advised to lower salt consumption, reported reduced sodium intake 12 months later (Nielsen and El-Sohemy 2014). In contrast, the personalised nutrition intervention had no effects on the intakes of caffeine, vitamin C, or sugars when participants were informed that they carried risk variants in the CYP1A2, GSTM1 and GSTT1, and TAS1R2 genes, respectively, perhaps because, as the authors suggested, baseline intakes of these specific food components were close to the recommended intakes so that participants had limited scope for dietary improvement (Nielsen and El-Sohemy 2014). A systematic review included in this Virtual Issue found four RCTs that investigated the impact of genetic variability on the relationship between caffeine and cardiometabolic outcomes (Virgili et al. 2023). However, differences in experimental protocols, in characteristics of the participants under study, in the genetic variants assessed, and in the chosen outcome measurements limit the conclusions that could be drawn (Virgili et al. 2023). The Food4Me Study tested the hypothesis that personalised nutrition advice is more effective than generic dietary advice in improving the overall healthiness of the diet and investigated the value of including data on phenotype and genotype, as well as baseline dietary pattern, in developing personalised nutrition advice (Celis-Morales et al. 2015). This internet-based RCT recruited and randomised 1607 European adults from seven different countries, of whom 1269 completed the 6 months study and provided estimates of dietary intake at both baseline and end-of-study (Celis- Morales et al. 2017). Those randomised to personalised nutrition treatments reported bigger improvements in overall dietary healthiness (measured as Healthy Eating Index) but there was no evidence that using more complex phenotypic and genotypic information when developing personalised nutrition advice was more effective than using baseline dietary pattern (Celis- Morales et al. 2017). Subsequent systematic reviews of the outcomes of relevant RCTs have concluded that personalised nutrition approaches produce modest but statistically significant improvements in dietary behaviours compared with generic advice, particularly when tailored to individual dietary intake and lifestyle factors (Jinnette et al. 2021; Misir et al. 2024). However, the added benefit of using genetic or phenotypic data as a basis for developing personalised nutrition advice remains uncertain. An analysis undertaken for the Food Standards Agency concluded that personalised nutrition is likely to remain niche for the foreseeable future, limiting the potential for broad impact on public health (Food Standards Agency 2023). In my view, this conclusion is unnecessarily pessimistic. With appropriate focus, and the deployment of novel technologies, notably Artificial Intelligence (AI)-based apps on smart phones, there is good potential to add personalised nutrition to the toolkit of cost-effective interventions designed to improve eating behaviour and to reduce the burden of diet-related disease. We need new tools in our toolkit because current one-size-fits-all approaches produce very modest improvements in eating behaviour and, in general, fail to address health inequities. Supermarkets have extensive knowledge of shoppers' behaviours and provide a key opportunity to use personalised nutrition approaches to shape individual food purchases and to improve eating patterns and health. This concept was tested in the SuperWIN trial in which researchers collaborated with a supermarket chain to conduct a supermarket and web-based, dietitian-led personalised nutrition intervention (Steen et al. 2022). Although the initial improvements in diet quality were not sustained, this study provides proof-of-principle for what might be possible. Future point-of-purchase interventions should take advance of AI-based tools that use complex and dynamic information on a wide range of relevant factors such as sociodemographic factors, health, psychosocial factors, food preferences, past purchasing behaviour and individual aspirations to deliver personalised and timely advice that can be actioned immediately. There is also considerable opportunity to develop analogous approaches to help individuals make healthier choices when eating out of home or ordering from meal delivery services. To be scalable and cost effective, such personalised nutrition interventions should not require active engagement from nutritionists or dietitians. Renner et al. (2023) have suggested similar ‘just-in-time AI systems’ for personalised nutrition that incorporate well-established behaviour change techniques. Personalised nutrition interventions of this kind are variants of so-called computer-tailored health communications (CTC). A recent systematic review and meta-analysis of outcomes from 11 RCTs involving this approach showed that they result in increased intake of fruits and vegetables, and the authors concluded that CTC is a feasible and efficacious way to promote sustained improvements in healthy eating habits and noted its relative affordability, minimal risks, and ease of implementation (Misir et al. 2024). To maximise benefits and to reduce the risk of exacerbating health inequity across populations, such individual-level interventions need to go together with extensive and continuous improvement in food environments (Pineda et al. 2022). Twenty-five years after the publication of the first comprehensive map of the human genome, the concept of personalised nutrition is now well-established, and current approaches are much broader than the early focus on genetics. Well-designed and conducted RCTs have demonstrated that personalised nutrition interventions can be superior to one-size-fits-all interventions in improving eating behaviour, but effect sizes remain relatively small (Jinnette et al. 2021). Despite, or perhaps because, the personalised nutrition field lacks clear regulatory oversight, defined standards, and consumer protection (Ordovas et al. 2018; Food Standards Agency 2023; Donovan et al. 2025) various personalised nutrition offerings are being commercialised, often without any published scientific evidence that they are effective. Although commercialisation may drive innovation and reduce costs of implementing personalised nutrition, it is probable that such commercial offerings will be taken up only by more affluent sections of society who have the resources to do so and who have interests in health or health technologies. This risks exacerbating inequity in dietary patterns and in health and may distract from the development of more inclusive personalised nutrition approaches that benefit all and, especially, those in greatest need of dietary improvement. From a public health perspective, future personalised nutrition interventions should focus more on psychosocial, cultural and economic factors and less on biology (Mathers 2019). There is a huge opportunity to use developments in AI to facilitate management and interpretation of the complex and dynamic information flows (including inputs from the user) that would underpin such personalised nutrition interventions and to deliver attractive, timely, and actionable advice and support that is, by definition, relevant for each individual receiving it. John C. Mathers is the sole author of this article. The author declares no conflicts of interest. Data sharing not applicable—no new data generated, or the article describes entirely theoretical research.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.